WorldThreatModelHarness

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.72). In Workflows/UpdateModels.md, the workflow invokes the Research/WebSearch skill to fetch and synthesize public web content at runtime, and that fetched outsider-authored text is then written into the world model markdown files that the agent later reads for TestIdea/ViewModels—creating an indirect prompt-injection path via stored model prose.